원문정보
초록
영어
This paper focuses on the path planning for a mobile robot which is operated in indoor environment. Since the layout of indoor environment is a hybrid structure of known and unknown, this paper presents a hybrid algorithm which uses the Max-Q method and the option method together. Firstly, a novel task graph and high level definition are presented to divide sub-tasks. Then, the appropriate definitions of states, actions and options could let a robot fulfill a task. Finally, an angle parameter is employed in the reward function to ensure a robot select a shorter path and adjust orientation timely. In the series of simulations, a robot can arrival any position successfully with random initial positions and directions. Moreover the results show that a robot can overcome the local minimal problem with our hybrid method.
목차
1. Introduction
2. Related Knowledge
2.1. Reinforcement Learning
2.2. Hierarchical Reinforcement Learning
3. The Design of the Hybrid Method
3.1. States and Actions Definition
3.2. Options Definition
3.3. Reward Function
4. Simulation
4.1. Partial Simulation
4.2. Whole Simulation
5. Conclusion and Future Work
References